I do what I love. It started with game development, then security, music for a while, and now AI. I build what I want to build, test it, and ship it in the open. I like it when people enjoy the things I make. These days I'm focused on AI agents and harnesses — figuring out how to make already-useful AI more convenient, safer, and more fun, and having a good time doing it.
제가 좋아하는 것을 합니다. 처음에는 게임 개발, 그 다음에는 보안, 한동안은 음악, 이제는 AI까지요. 만들고 싶은 걸 만들고, 테스트하고, 공개합니다. 제가 만든 것을 남들이 좋아해주는 것을 좋아합니다. 지금은 AI 에이전트와 하네스에 집중하고 있습니다 — 이미 충분히 유용한 AI를 어떻게 더 편리하게, 더 안전하게, 더 재미있게 만들 수 있을지 고민하며, 즐겁게 개발하는 중입니다.
- AI Agent Security — Defense benchmarking and automated evaluation across proprietary APIs (OpenAI, Anthropic, Google) and open-weights models (DeepSeek, Qwen, Gemma, Llama), grounded in hands-on prompt injection research (write-ups in preparation)
- Agent Engineering — Anthropic/OpenAI Agents SDK, MCP (Model Context Protocol) server development and tool integration
- Evaluation Infrastructure — Automated evaluation harnesses for LLM defenses and adversarial attack tracking
- ML Pipelines — Stable Diffusion LoRA training, computer vision model fine-tuning
Languages & Tools: Python, TypeScript (primary) · Java, Go, Rust, C++ · FastAPI, Spring Boot, Node.js · Docker, Kubernetes
Security Domains: Web vulnerability assessment (secure coding, SAST/DAST) · Digital forensics (artifact & system analysis)
- music-composition — Modular music-composition skill for AI assistants: harmony, melody, form, 24+ genres.
- Antikythera — Python-first local math MCP server; a practical, low-cost alternative to Wolfram-style helpers for AI agents.
- superpowershell — An agent-friendly Windows PowerShell environment guide for humans and local coding agents.
- HanjaKey — macOS menu-bar app that revives the Korean Hanja key: Hangul→Hanja, KS X 1001 symbols, and Hanja-word conversion via a global hotkey, inserted in place.
- rob-a-bank — Standalone reimplementation of my prompt injection wargame challenge originally authored for the HAF platform (original source lost; this will be a faithful rebuild, clearly marked as such).
- Wish (Wish Is Self-sharpening Harness) — an AI agent harness that adapts to you instead of asking to be configured. Self-tuning defaults, zero-config onboarding. Currently in design.
Enterprise cloud security & application modernization (AM) program.
Next-generation white hat hacker academy hosted by the Ministry of Science and ICT.
- Core member of HAF, the CTF/wargame engineering team
- Authored and deployed "Rob A Bank" — a prompt injection wargame challenge evaluating system prompt vulnerabilities on the HAF platform
- Top Excellence Award (2nd Place), Team Category
- Encouragement Award (7th Place), Individual Category
Completed the government-funded gifted education program (Basic Track II) with an Academic Excellence Award.
- High School Graduation Equivalency Examination — Passed
- Sunduck High School (Autonomous Private High School) — left to focus on developing career-relevant skills and personal research projects via the equivalency track
한국어 상세 이력 펼치기
- LG CNS AM Inspire Camp 5기 (2026.05 – 현재) — 엔터프라이즈 클라우드 보안 및 AM 실무 과정
- 화이트햇 스쿨 2기, KITRI (2024.03 – 2024.09) — 과학기술정보통신부 주관 차세대 보안리더 양성 프로그램 수료
- CTF 워게임 제작·운영 팀 HAF 핵심 멤버
- 시스템 프롬프트 취약점 검증용 프롬프트 인젝션 워게임 "Rob A Bank" 출제 및 개발
- 제9회 교육부 정보보안 경진대회 (2023.11) — 단체전 2위(최우수상), 개인전 7위(장려상)
- 서울여자대학교 정보보호영재교육원 (2023.03 – 2023.12) — 중등기초II반 우수 수료
- 고졸 검정고시 합격
- 선덕고등학교 (자율형 사립고등학교) — 진로에 맞는 역량 향상 및 개인 연구 프로젝트에 집중하기 위해 검정고시 경로 선택
- AI/에이전트: Anthropic·OpenAI Agents SDK, MCP 서버 구축 및 도구 연동, LLM 방어 자동화 평가 하네스, 주요 상용 API 및 오픈 가중치 모델 대상 정렬 검증·레드티밍, Stable Diffusion LoRA 학습, CV 모델 파인튜닝
- 개발/DevOps: Python, TypeScript (주력) · Java, Go, Rust, C++ · FastAPI, Spring Boot, Node.js · Docker, Kubernetes
- 보안: 웹 취약점 분석(시큐어 코딩, SAST/DAST), 디지털 포렌식(아티팩트·시스템 분석)
2026-08-19— Pushed commits to SJY051/D32026-08-19— Pushed commits to SJY051/D32026-08-19— Pushed commits to SJY051/D32026-08-19— Opened PR ? in SJY051/D32026-08-19— Pushed commits to SJY051/D3
Auto-generated by a scheduled GitHub Action — profiles should maintain themselves.
- Email:
rt.kirby2009@gmail.com - Open an issue for collaboration, research discussions, or code review — always welcome.